The Problem Solution Use Cases Technology Careers Get in Touch
Tiv Language AI

Tiv deserves AI that
actually understands it.

Today's tools treat Tiv as an afterthought by flattening its tones, guessing at meaning, with no voice and no real-world products. Kwaghfan is building an open, community-sourced foundation for Tiv AI, and the applied tools that put it to work.

In development. We are designing the corpus and model.
5M+ Tiv speakers
0 Production voice models for Tiv
Open To be community-built & openly licensed
Where we are

Early, and building carefully

Kwaghfan is at the design stage. Nothing is deployed yet. We are taking the time to get the data right, because every model we build later depends on it.

Now

Reviewing the research

We are reading what has already been done for Tiv and for other low-resource African languages, so we build on that work instead of repeating it.

Now

Designing the collection protocol

We are designing how we will collect Tiv text and speech so the corpus captures the right data: dialects, everyday registers, tone, and clear consent from every contributor.

Now

Choosing the model approach

We are deciding how to build the model from that data, starting from open-source multilingual models that can run on modest hardware.

Next milestone

A data-collection pilot in Benue, where a small team of native speakers will record, transcribe, and review a first batch of Tiv data. We are hiring for it.

The Problem

Tiv AI exists. It isn't good enough.

A handful of tools can now translate Tiv text. But they treat the language as a low-resource afterthought, flattening the tonal distinctions that change meaning, and hallucinating on anything legal, medical, or official. They're unreliable exactly where accuracy matters most.

There is no reliable voice. Speech recognition and text-to-speech for Tiv remain perpetually "coming soon," because the phonetic data needed to build them doesn't exist yet.

And where good data does exist, it's locked away as proprietary corpora built to be sold, not shared. Nobody has built the applied products either.

0 Production-ready voice models for Tiv. Speech recognition and synthesis remain unsolved.
Our Approach

A foundation built the right way

Kwaghfan is building Tiv AI from the ground up, open where others are closed, voice-capable where others stop at text, and deployed where others ship only an API.

An Open Tiv Corpus

A community-sourced dataset collected from inside Tiv-speaking communities, designed to cover dialectal variation and real conversational registers. It will be openly licensed and shared, not sold.

Voice-First Models

Beyond text, we plan to build tone-aware speech recognition and synthesis that sounds natural to Tiv speakers.

Deployed, Applied Products

We plan to release a free public chatbot and tools for health, agriculture, and government, built for people in Benue to use in the language they think in.

Use Cases

Real applications, real impact

Agriculture

Agricultural Extension

Planting calendars, pest management, subsidy information, and best practices reaching farmers in the language they think in, not the language of bureaucracy.

Government

Government Services

Voter education, civil registration guidance, policy communication, and public announcements making government accessible to every citizen, not just English speakers.

Education

Literacy & Learning

Supporting Tiv-language education, enabling students to access learning materials in their mother tongue, and preserving the language in the digital age.

NGOs

Development Programs

International development organizations serving Benue State, such as WHO and UNICEF programs, could get a communication layer that reaches the people they serve.

Technology

How Kwaghfan will work

The planned system uses a bridge architecture that combines a purpose-built Tiv language model with large-scale AI reasoning, designed to be reliable without requiring massive compute. Voice comes next, layered on the same foundation.

1

User speaks Tiv

A Tiv speaker will type or speak their question in natural Tiv through the chat interface.

2

Tiv model translates

A fine-tuned Tiv model will convert the input to English and extract the user's intent.

3

AI reasons in English

A large language model will answer the query using domain-specific knowledge bases.

4

Response in Tiv

The answer will be translated back into natural Tiv and returned to the user.

Community-sourced data

We are designing a community competition to collect Tiv text across health, agriculture, commerce, and daily life, written by Tiv speakers.

Open-source foundations

We plan to fine-tune open-source multilingual models with parameter-efficient methods, so there is no proprietary lock-in and the models can run on modest hardware.

Open by design

Designed for open release from day one. Data will be verified by native speakers, documented to research standards, and licensed to be shared. A public good, not a walled garden.

Collaboration

Built for partnership

We are looking for partners to shape the corpus and first deployments. If your organization needs to reach Tiv-speaking communities, we would like to hear what you need.

Government Agencies

Help us choose the first deployments, so what we build matches the programs you run and the questions citizens ask.

NGOs & Development

Tell us which health, education, and agricultural information your programs need to deliver in Tiv, and help us test early versions with the communities you serve.

Research Institutions

Collaborate with us on the corpus design and collection protocol, and on expanding NLP resources for African languages.

Technology Companies

Support the work with compute, tooling, or expertise in low-resource speech and language models.

Bring AI to every language

Whether you're a government agency seeking to reach Tiv speakers, an NGO deploying programs in Benue State, a researcher interested in low-resource NLP, or an investor in African AI infrastructure, we'd like to hear from you.